Can ChatGPT Make Videos in 2026? What It Can and Can't Do
ChatGPT writes the script, shot list and prompts, but the clip comes from a video model. Two routes to get from a chat to a finished video, with credit costs.
No: as reported, ChatGPT does not generate video itself. The Sora app was reported shut down on 26 April 2026 and the Sora API on 24 September 2026, and video generation is not available in ChatGPT, though one source disagrees on the details. What ChatGPT does well is the writing around the clip: the script, the shot list and the prompts. A video model then makes the footage, and 5 seconds of 720p Seedance 2 Fast costs 50 credits.
Updated October 2026.
Because the status can change, confirm the current position in OpenAI’s help centre before you build a workflow on it. Everything below is designed to work either way.
Short answer: what ChatGPT does itself and what it hands off
Treat ChatGPT as the pre-production department. It can plan a video, write a script with timing, break the script into shots, describe camera moves and write the prompt that a video model reads. It does not render frames, does not hold a video length selector, and does not output an MP4.
What it hands off is the generation itself. That handoff is the only part that changes between route one and route two below. In route one you copy the prompts into a video generator yourself. In route two you connect a generation service to the assistant and ask for the clip from inside the chat.
What ChatGPT is good for in a video job: script, shot list, prompts
The useful output is four artefacts, each with a job.
- A spoken script. One idea per sentence, timed to the clip length you can actually afford. A 10-second clip holds about 25 spoken words, not a voice-over essay.
- A shot list. One row per shot, with duration, subject, action, camera and setting.
- Generation prompts. One prompt per shot, written in the order a video model reads: subject, action, camera, setting, light, style, sound.
- A review list. The three things to check on the first render: identity, hands, and any on-screen text. Ask for that list up front and the second attempt is faster.
The wider craft of getting a usable script out of a language model is covered in the AI video script writer guide. If you want a pre-built workflow for turning a written script into footage, the MCP route is documented separately.
Route 1: paste ChatGPT’s prompts into a video generator
This is the route with no setup. Ask the assistant for a shot list, copy one prompt at a time into text to video or image to video, choose the model and length, and read the price before you generate.
- Fix the destination first. 9:16 for Reels, Shorts and TikTok; 16:9 for a site hero or a YouTube pre-roll. Set the same aspect ratio in the prompt and in the tool.
- Pick a clip length you can afford. Seedance 2 Fast runs 4 to 15 seconds in one-second steps; Seedance 2.5 goes to 30 seconds. Long clips punish vague prompts, so start short.
- Generate the cheapest useful draft. Seedance 2 Fast at 720p for 5 seconds is 50 credits. Seedance 2.5 at 720p for 5 seconds is 126. Veo 3.1 Fast is a flat 40 credits per clip with no duration selector.
- Judge one thing at a time. If the motion is right but the face drifts, keep the prompt and change only the identity wording or the reference image.
- Finish the keeper at the delivery settings. Re-run the winning prompt at the resolution and length you will publish, and do not re-run the losers.
Keep the shot list next to the renders. When shot three fails, you want the exact prompt and settings, not a memory of them.
Route 2: connect your assistant over MCP and ask for the clip
VdoBloom runs an MCP server at https://vdobloom.com/mcp that an assistant can call. In Claude or ChatGPT you add it as a custom connector and sign in, with no key to paste. Command-line clients can add the same server with a key in an authorization header. This is a custom connector, not an official OpenAI feature.
Once connected, the assistant can list models, check a balance, generate an image, generate a video from text or from an image URL, and poll a job. The same credits and the same prices apply as in the app, one-time pack credits never expire, and a failed generation is refunded. Generation takes one to ten minutes, so the call returns a job. Full setup, tool names and limits are in the MCP walkthrough, and the API surface is on the developer page.
The gain is that the shot list and the generation stay in one conversation: ask for a 5-second 9:16 clip, then ask for the same shot at a different length. The limit is that MCP exposes generation, not the specialist tools; effects, lip-sync, avatars and the ad wizards stay in the dashboard.
Three prompts to copy: script, shot list, video prompt
These are prompts to paste into ChatGPT, not prompts that a video model reads. Each one is deliberately narrow.
- Script prompt: “Write a 10-second spoken script for a vertical video about a compact desk lamp. One idea per sentence, no more than 25 words in total, no brand name, end on the product in use.”
- Shot list prompt: “Turn that script into a shot list. One row per shot with columns: shot number, duration in seconds, subject, action, camera move, setting and light. The total must be 10 seconds. One action per shot.”
- Video prompt: “Write a single generation prompt for shot two. Order: subject, action, camera, setting, light, style. Name the subject once. End with the sound. Do not mention anything already visible in the starting image.”
Ask for the prompts one shot at a time. A batch of six prompts invites the assistant to reuse the same phrasing, and identical phrasing across shots is how characters drift between clips.
Can ChatGPT edit an existing video?
Not by itself. As reported, ChatGPT does not expose video generation or video editing, so it cannot cut, restyle or re-voice a clip you already have. It can write the instruction, describe the change and review a description of the result, but the edit needs a video-to-video model.
VdoBloom’s AI Edit tab runs FLUX Video Edit [fast], which takes a clip of 0.7 to 15 seconds and one written instruction: swap an object, change the setting, restyle the clip or change what someone says. The source clip sets the length, framing, camera, timing and audio. For anything longer than 15 seconds, cut the section first, edit it, and rejoin. That is a different workflow from generation, and it is covered in the FLUX Video Edit guide.
Which model for which kind of clip
| Job | Model to start with | Credits |
|---|---|---|
| Cheap prompt testing | Seedance 2 Fast, 720p, 5 s | 50 |
| One long continuous take | Seedance 2.5, 720p, 5 s | 126 |
| Fast single clip, fixed length | Veo 3.1 Fast | 40 |
Seedance 2.5 runs 4 to 30 seconds, six aspect ratios including 9:16 and 21:9, at 480p, 720p or 1080p, with audio generated by default. Its model page is here. Veo 3.1 Fast is a flat 40 credits with no duration selector and 16:9 or 9:16 only. Every exact price is shown before you generate, and plan and pack options sit on the pricing page. For a wider comparison of what to use when the assistant is not the bottleneck, see Sora alternatives.
Frequently asked questions
Can ChatGPT make a video in 2026?
As reported, no. The Sora app and API were reported shut down in 2026, and video generation is not available in ChatGPT; one source disagrees, so check OpenAI’s help centre.
Can I connect VdoBloom to ChatGPT?
Yes, through a custom connector that points at the VdoBloom MCP server and signs in. It is not an official OpenAI feature, so you add it in the connector settings rather than in ChatGPT’s core tools.
What should I ask ChatGPT to write for a video?
A timed spoken script, a shot list with one action per shot, and one generation prompt per shot in the order subject, action, camera, setting, light, style, sound.
Can ChatGPT edit a clip I already have?
As reported, no. Use a video-to-video model for the edit. FLUX Video Edit [fast] takes 0.7 to 15 seconds and one instruction, and keeps the source timing and audio.
How much does the first clip cost?
A 5-second 720p Seedance 2 Fast clip is 50 credits and a Veo 3.1 Fast clip is a flat 40. New accounts start with 10 free credits and no card.
Write the script in your assistant, then make the clip in text to video or from a still in image to video. New accounts start with 10 free credits and no card, and one-time packs never expire, so a first 5-second test does not commit you to a plan.
Ready to try it?
Create your first AI video in minutes — no credit card required.
Start Creating Free →